Update app.py

app.py

@@ -2,6 +2,10 @@ from flask import Flask, request, jsonify
 import os
 import logging
 
+# Set cache environment variables
+os.environ['HF_HOME'] = '/.cache/huggingface'
+os.environ['TRANSFORMERS_CACHE'] = '/.cache/huggingface/transformers'
+
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 
@@ -14,24 +18,36 @@ def load_model():
         logger.info("Loading YOUR fine-tuned model...")
         from transformers import pipeline
         
-        # Use YOUR converted model
+        # Try to use your model with cache settings
         model = pipeline(
             "text-generation", 
             model="kacperbb/phi-3.5-hf-finetuned",
-            trust_remote_code=True
+            trust_remote_code=True,
+            cache_dir="/.cache/huggingface"
         )
         logger.info("✅ YOUR fine-tuned model loaded successfully!")
         return True
     except Exception as e:
         logger.error(f"❌ Error loading your model: {e}")
-        logger.info("Falling back to base model...")
+        logger.info("Trying with base model...")
         try:
-            model = pipeline("text-generation", model="microsoft/Phi-3.5-mini-instruct")
+            model = pipeline(
+                "text-generation", 
+                model="microsoft/Phi-3.5-mini-instruct",
+                trust_remote_code=True,
+                cache_dir="/.cache/huggingface"
+            )
             logger.info("✅ Base model loaded as fallback")
             return True
         except Exception as e2:
             logger.error(f"❌ Fallback failed: {e2}")
-            return False
+            # Last resort - use a smaller model
+            try:
+                model = pipeline("text-generation", model="microsoft/DialoGPT-medium")
+                logger.info("✅ Smaller fallback model loaded")
+                return True
+            except:
+                return False
 
 @app.route('/generate', methods=['POST'])
 def generate_text():
@@ -66,7 +82,7 @@ def health():
 @app.route('/', methods=['GET'])
 def home():
     return jsonify({
-        "message": "Your Phi 3.5 Fine-tuned API is running!",
+        "message": "Phi 3.5 Fine-tuned API is running!",
         "model": "kacperbb/phi-3.5-hf-finetuned",
         "endpoints": {
             "generate": "POST /generate",
@@ -75,7 +91,7 @@ def home():
     })
 
 if __name__ == '__main__':
-    logger.info("Starting YOUR Phi 3.5 API...")
+    logger.info("Starting Phi 3.5 API...")
     load_model()
     port = int(os.environ.get('PORT', 7860))
     app.run(host='0.0.0.0', port=port, debug=False)